Introduction

Since the end of the Cold War, most authoritarian countries have held national elections that allow opposing parties to compete for and win seats. Although such authoritarian elections are often extensively rigged and do not allow the possibility of government turnover, growing pressure from the international community makes it difficult to hold periodic elections without the participation of opposition parties (Norris 2014: 78–81). This post-Cold War proliferation of “electoral authoritarianism” (Schedler 2002) or “competitive authoritarianism” (Levitsky and Way 2010) revived the study of authoritarian politics in the field of comparative politics. Nevertheless, the conditions under which electoral authoritarian regimes democratize have yet to be fully understood. Some electoral authoritarian countries, such as Malaysia (1974–2018), Singapore (1965–present), Zimbabwe (1980–2018), and Egypt (1979–2011), are examples of dictators who successfully contained the pressures of democratization. In contrast, other electoral authoritarian regimes, such as those in the Philippines (1978–1986), Kenya (1992–1997), Guatemala (1955–1957), and Honduras (1954–1956), collapsed relatively quickly after establishment and were followed by democratic transitions. How can these variations be explained?Footnote 1

While the burgeoning literature on this question focuses on various types of political institutions, we focus on one type that is not yet adequately scrutinized: executive selection systems. An executive selection system refers to the formal (law-based) rule regarding the selection of a polity’s chief executive. A fundamental characteristic of this system is whether the executive is elected directly by voters or indirectly by a designated body, typically a parliament. Following Roberts (2015), we call the former a Presidential system and the latter a Parliament-based system. Juan Linz famously argued that Presidential systems are peril to democracy by suggesting that they are more prone to break down than Parliament-based systems (Linz 1990a). In this article, we argue that in the context of transitions from electoral authoritarianism to democracy, Parliament-based systems are less likely to democratize than those with Presidential systems; hence, Parliament-based systems can be a peril for those forces seeking democratization.

We argue that Parliament-based electoral authoritarian regimes are less likely to democratize because they bundle the following three mechanisms: First, Parliament-based systems facilitate credible power-sharing among ruling elites, often incentivizing dictators to have a highly institutionalized ruling party. This system can effectively prevent democratic transitions by deterring ruling elites’ defections from the regime. Second, a surprise win for the opposition is more likely in elections under Presidential-based systems when a charismatic leader emerges. In Parliament-based systems, the popularity of a charismatic opposition leader only has an indirect effect over the elections in districts other than their own, as most parliamentary elections have multiple electoral districts. Third, parliamentary dictators retain much more room to arbitrarily redistrict and apportion according to the interests of dictators and their parties compared with presidential elections. These indirect measures of electoral manipulation allow dictators to enjoy bias in parliamentary seats to maintain an overwhelming majority, or in some cases win the majority of seats with less than a majority of the votes, thus preventing democratization through elections.

To test our argument empirically, we conduct several cross-national statistical analyses. Instrumental variables estimation finds that electoral authoritarian regimes with Parliament-based systems are less likely to democratize than those with Presidential systems. This statistical result remains robust across a battery of sensitivity analyses. Furthermore, we provide additional cross-national evidence on causal mechanisms: Parliament-based systems are more likely to maintain election victories through indirect electoral manipulation and have institutionalized ruling parties.

This study contributes to the literature on democratization, authoritarian politics, and political institutions. We empirically show that the executive selection systems are likely to affect democratic transitions. The distinct characteristics of presidentialism and parliamentarism have given rise to a debate among scholars regarding their differing effects on democratic breakdown (Linz 1990a, 1990b; Lijphart 1991; Horowitz 1990). We establish the distinction between Parliament-based and Presidential systems as an important factor influencing regime transition from electoral authoritarianism to democracy, arguing that Parliament-based systems discourage electoral autocracies from democratizing and thus can be a peril for democratization.

More specifically, our study expands that of Roberts (2015) on authoritarian durability in a couple of important ways. First, we employ an alternative dependent variable that directly addresses democratic transition. While our main independent variable is the same as that of Roberts, the probability of democratization is our main dependent variable, whereas Roberts’s was, broadly, autocratic breakdown, including a change to a new authoritarian regime. Second, we suggest and empirically test a mechanism that influences the likelihood of democratic transition by elections: Presidential systems tend to experience blatant electoral fraud more frequently than Parliament-based systems presumably because of the presence of indirect electoral manipulation such as gerrymandering and malapportionment. We also provide direct evidence for Roberts’s insight that Parliament-based systems lead to better power-sharing capabilities by being able to measure ruling party institutionalization. Third, this study covers a longer time period (1946–2012) than Roberts’s (1975–2012). Lastly, to deal with the possibility of omitted variables’ bias and reverse causality, we employ instrumental variable (IV) estimators. By dealing with unobservable heterogeneity and possible reverse causality, IV estimators allow for robust estimates of causal relationships between executive selection systems and democratic transition.

Executive Selection Systems and Democratic Transitions

Two broad classes of arguments have been advanced to explain the transition from electoral authoritarianism to democracy. The first focuses on international factors. Scholars have emphasized the role of “international diffusion” (Brinks and Coppedge 2006; Gleditsch and Ward 2006; Huntington 1991), through emulation and demonstration effects, as well as foreign intervention, such as democracy promotion aid, aid conditionality, and electoral monitoring (Donno 2013; Kelley 2012; Pevehouse 2002).Footnote 2 The second broad class of arguments pertains to domestic politics. Numerous studies analyze the impact of socio-economic structure, such as the level of modernization (Lipset 1959; Przeworski et al. 2000), natural resource wealth (Ross 2012), and economic inequality (Boix 2003). Some scholars focus on opposition strategies (Howard and Roessler 2006; Bunce and Wolchik 2009; Donno 2013) and others emphasize the role of institutions. For example, dominant parties consolidate their rule (Geddes 1999; Brownlee 2009) by facilitating power-sharing among regime elites (Magaloni 2008; Svolik 2012; Geddes et al. 2018). The presence of a multi-party legislature encourages the co-option of the opposition through credible policy concessions (Gandhi 2008; Gandhi and Przeworski 2007). Elections can allow the citizenry to engage in economic distribution (Higashijima 2022; Blaydes 2011; Magaloni 2006), gather information on popular sentiments (Miller 2015a; Reuter and Robertson 2012), and even divide the opposition (Lust-Okar 2004). Further, adopting constitutions (Albertus and Menaldo 2012) and creating institutionalized succession rules (Frantz and Stein 2017) can facilitate elite inner circles to remain loyal to the authoritarian leader.

This study engages with the domestic-institutionalist strand of the literature and explores the role of the selection method for the chief executive. Specifically, we argue that, other things being equal, electoral authoritarian regimes that adopt Parliament-based systems for executive selection are less likely to democratize than those adopting Presidential systems. As a reminder, in a Parliament-based system, the chief executive (usually called prime minister) is elected indirectly, usually by the legislature,Footnote 3 whereas in a Presidential system, voters in most cases directly elect the chief executive (president). The Parliament-based system is equivalent to parliamentarism and the Presidential system (and the semi-presidential system) is equivalent to presidentialism regarding the election method for the chief executive. Since we are concerned with the selection method of the chief executive, we include pure-presidential and semi-presidential systems in the category of Presidential systems; in both systems, the chief executive is not elected by the parliament. We expect that these constitution-level differences create institutional conditions that influence the prospects of democratization for the following three mechanisms.

First, Parliament-based systems are more conducive to institutionalizing the ruling party’s organization. An institutionalized political party has a well-structured organization, and its members are loyal to its goals and decisions (Huntington 1968; Mainwaring and Scully 1995). Although not exactly adopting the concept of institutionalization, many have indicated that direct election systems tend to have less institutionalized parties than parliamentary systems (Linz 1990b: 89; Shugart 1998: 2). For example, Samuels and Shugart (2010) convincingly demonstrate that presidential and semi-presidential systems tend to produce less institutionalized parties than parliamentary governments. When the chief executive is directly elected, parties tend to become presidentialized. In such parties, according to Samuels and Shugart, party organizations and activities are strongly influenced to maximize chances of winning presidential elections (Samuels and Shugart 2010). This is accomplished by pursuing vote-maximizing policies at the cost of promoting the party’s ideological positions and by concentrating party resources on winning presidential elections rather than on building the party as a whole. In combination, these features hinder the institutionalization of parties—both ruling and opposition—in Presidential systems.

When the ruling party in an electoral authoritarian regime is under-institutionalized, the country is more likely to experience regime instability because an institutionalized governing party serves as a mechanism for credible power-sharing among the ruling elites (Magaloni 2008; Svolik 2012).Footnote 4 Under such circumstances, the elites within the ruling circle can expect to fulfill career ambitions and benefit from belonging to the regime in the future. Without such an institution, the regime elites have stronger incentives to defect to the opposition and more interest in staging a palace coup to oust the ruler by force. Simultaneously, when the party is not institutionalized, leadership succession may not occur according to party rules. Thus, the chief executive’s resignation, death, or retirement is more likely to cause a violent political disturbance. Additionally, an under-institutionalized ruling party cannot facilitate the efficient distribution of patronage to voters, which makes it difficult for the party to win by significant margins in elections (Greene 2010; Magaloni 2006).

A lack of ruling party institutionalization and subsequent party personalism may encourage violent incidents and thus regime instability (Geddes et al. 2018), which is an important background condition for democratization (Miller 2021). However, the existence of only this mechanism does not necessarily bring democratic transitions. The following two mechanisms bundled by executive selection systems are also important. The second mechanism concerns the ease of opposition coordination under Presidential systems, or what Sato and Wahman (2019) call lateral threats. Namely, where a charismatic opposition leader emerges, Presidential systems make it easier for them to win, and thus to democratize, than Parliament-based systems.Footnote 5 This arises due to several reasons. The first reason is related to our first point about the party institutionalization. Since the ruling party in Presidential-based systems tends to be less institutionalized than its Parliament-based counterparts, voters as well as political machine brokers are more likely to change their allegiance to the charismatic opposition leader. Further, the weakness of ruling party organization makes it harder for autocrats to obtain political information from the grass-roots level in order to calculate the probability of his victory in the next election. This can lead to misallocation of state resources for electoral purposes, and/or setting the election timing under an unfavorable political environment.Footnote 6 The second reason is that for the opposition to take power, a single victory in a nationwide constituency is all what it takes in Presidential systems,Footnote 7 while a slate of candidates from the same party coalition has to win seats to form a majority in Parliament-based systems. Recruiting a whole slate of candidates that are appealing enough to voters to obtain the majority seats is harder than promoting a single super star candidate to win a majority vote in a single race. The popularity of a charismatic opposition leader might create some “coattail effects” to his fellow candidates, yet it is likely to be only marginal. In addition, even when the ruling party in Parliament-based system loses a majority, it can still form a coalition with another party thereby avoiding the loss of executive power (Roberts 2015, p.926). Such post-election maneuver is not possible in Presidential systems. For these reasons, Presidential systems provide a greater chance of democratization when a charismatic opposition leader emerges than Parliament-based systems.Footnote 8

Third, autocrats in Parliament-based systems have a wider variety of electoral maneuvering tactics available than those in Presidential systems. In studies of electoral malpractice, Birch and Schedler each observe three main types of malpractice: electoral rules, vote choice, and electoral administration (Birch 2011; Schedler 2013). In the manipulation of electoral rules, parliamentary autocrats enjoy an advantage over their presidential counterparts. Particularly, prime ministers have two additional manipulation strategies available. The first is malapportionment: a disproportional allocation of legislative seats considering the number of people in the electorate. The ruling party can apportion a larger number of seats to their bailiwicks to win a greater number of seats with fewer votes.Footnote 9 In particular, research suggests that legislative malapportionment tends to be frequently used as an electioneering strategy in “mixed regimes” like electoral authoritarian countries (Ong et al. 2017). The second is gerrymandering: drawing district boundaries in a politically calculated manner. Typically, the ruling party can minimize the number of opposition seats by packing opposition supporters into districts where the ruling party is certain to win and/or splitting the opposition’s stronghold into several districts so that the opposition loses (Lust-Okar and Jamal 2002). Although malapportionment and gerrymandering are possible in Presidential systems’ legislative elections, these elections do not affect the choice of chief executive. Accordingly, autocrats cannot use these indirect manipulation techniques when electing a president. One precondition for such strategies is that legislative elections do not use proportional representation (PR) with a single nationwide constituency. Technically, in such an electoral system, neither form of wrongdoing can occur. However, few electoral authoritarian regimes use a nationwide PR system.Footnote 10

Malapportionment and gerrymandering can prevent democratization in several ways. First, they can facilitate the appearance of an overwhelming victory by the ruling party without popular backlash because these methods of electoral manipulation are covert and less visible and the percentage of seats it acquires is disproportionately larger than the percentage of votes in the case of highly malapportioned and/or gerrymandered elections (Ong 2018; Higashijima 2021). Second, they allow for a “spurious majority,” referring to election results wherein one party wins a majority of seats while another party wins the majority of votes (Siaroff 2003). For example, in the case of Malaysia’s 2013 election, the opposition coalition Pakatan Rakyat won the majority of the popular votes (51%). However, the ruling coalition UMNO (United Malays National Organization), which won 47% of the votes, was able to maintain the majority status by occupying 133 seats in the 222-seat lower house. Conversely, in Presidential systems, chief executives cannot resort to such garden variety manipulation, thus tend to engage in blatant electoral fraud, which is likely to cause mass mobilization calling for democratization, or what Sato and Wahman (2019) call vertical threats. The Philippine’s 1986 election is a case par excellence of this type of transition: Marcos’s excessive electoral fraud triggered a mass demonstration, eventually forcing Marcos to flee the country.

We argue that the combination of these three mechanisms creates a condition that increases the probability of democratization in Presidential systems in comparison to Parliament-based systems.Footnote 11 In Presidential systems, the ruling party is less institutionalized and thus autocrats are more likely to face regime instability. Meanwhile, Presidential systems also allow the coordination of the opposition and increase the need to manipulate election results by blatant electoral fraud, which often provoke popular protests for democratic transitions. Conversely, in Parliament-based systems, the ruling party is highly institutionalized enough to maintain regime stability; parliamentary elections decrease the likelihood of opposition coordination and allow autocrats to use covert institutional manipulation like gerrymandering and malapportionment, contributing to preventing democratization. Note that our emphasis here is the likelihood of democratization. We expect that when all the three mechanisms are combined under a particular executive selection system, regime transition outcome is likely to be a democracy, but not necessarily transitions from one autocracy to another — autocratic transitions.

This study complements the existing studies on the link between the executive selection system and democratic transitions. In particular, in a recently published article, Roberts (2015) examines a similar argument, exploring the hypothesis that authoritarianism in a Presidential system is shorter-lived than in its parliamentary counterparts.Footnote 12 However, Roberts’s analyses differ from ours in several important respects. First, Roberts focuses on autocratic breakdown in general, whereas our analyses specifically investigate the transition from autocracy to democracy. Roberts’s main dependent variable is whether “the incumbent or his designated successor is removed from office via elections or violent means such as a coup or rebellion, or when the ruling party markedly changes the formal or informal rules for choosing leaders” (Roberts 2015: 527), including transitions from autocracy to democracy as well as those from one authoritarian regime to the next. In our analyses, in contrast, we expect that transition to democracy is predicted by executive selection systems due to the three mechanisms we discussed above.

Second, and related to the first point, we newly identify and empirically test an additional mechanism that influences the likelihood of democratic transition: Presidential systems tend to experience blatant electoral fraud more frequently than Parliament-based systems, presumably as the result of the absence of indirect electoral manipulation tools such as gerrymandering and malapportionment. In addition, we provide another empirical test to Roberts’s claim that Parliament-based systems have better power-sharing capability as the main mechanism behind his argument. While Roberts’s analyses mainly focused on the constraints on the chief executive, we analyze the relationship between the executive selection methods and the emergence of institutionalized ruling parties.

Third, to empirically test the validity of our argument and Roberts’s (2015), our analyses cover a longer period (1946–2012) than Roberts’s (1975–2012). To better identify the causal relationship, we also use instrumental variable estimators to deal with likely problems of endogeneity between executive selection systems and democratic transitions. Consistent with our theoretical expectation, our analysis reveals that utilizing the more extensive dataset, executive selection systems better explain democratic transitions than autocratic breakdown (Appendix C and Table C-1).Footnote 13

Our argument parallels the “perils of presidentialism” debate on the stability of democracy in electoral authoritarianism settings: whether parliamentarism leads to stability while presidentialism is prone to democratic breakdown (Linz 1990a, 1990b; Lijphart 1991; Horowitz 1990). Empirical analyses have not yielded conclusive results, with some showing parliamentary stability (Przeworski et al. 2000; Stepan and Skach 1993) and others showing that presidentialism per se is not the cause of democratic breakdown (Cheibub 2007), instead showing that it is the product of its combination with other political factors (Mainwaring 1993; Shugart and Carey 1992). Our argument suggests that executive–legislative relations (and, particularly, the method of executive selection) may affect not only democratic regimes’ stability as argued by scholars such as Linz (1990a, 1990b), Lijphart (1991), and Horowitz (1990) but also democratic transitions from electoral authoritarian regimes. We posit that institutions that regulate executive–legislative relations exert a similar influence in the context of limited (but not entirely restricted) electoral competition. It should be noted that we expect these effects to be seen only in electoral authoritarianism because we expect that the three mechanisms above, emanating from the difference in executive selection systems, occur only in the presence of some degree of electoral competition. By definition, in closed authoritarian regimes, no electoral competition exists in which a bona fide opposition has a chance of winning a seat.

Cross-National Statistical Analysis

Regime Type and Dependent Variable

To test our hypothesis, we conduct statistical analyses that cover 93 electoral authoritarian countries between 1946 and 2012. Following recent studies on authoritarian regimes, we assume that there are at least two subtypes of the authoritarian regime with respect to their level of electoral competition. The first is a closed authoritarian regime without meaningful electoral competition.Footnote 14 The second subtype is an electoral authoritarian regime with a level of electoral competition severely limited in favor of the ruling party. Adding the democratic regime classification yields three categories: (1) democracy, (2) electoral authoritarianism, and (3) closed authoritarianism.

To distinguish democracy from autocracy and identify democratic transitions, we use Geddes et al.’s (2014) definition of political regimes. According to them, democratization happens when “a competitive election for the executive, or for the body that chooses the executive, occurs and is won by a person other than the incumbent or someone allied with the incumbent; and the individual or party elected is allowed to take office.”Footnote 15

Then, among authoritarian regimes, we distinguish electoral autocracy from closed autocracy. To do this, researchers construct a dummy variable for the presence of limited multi-party competition in authoritarian regimes (Brownlee 2009; Svolik 2012) based on a variable measuring the degree of electoral competition in the World Bank’s Database of Political Institutions (DPI).Footnote 16 According to their definition, a country is an electoral authoritarian regime if it allows multiple parties to participate in elections.Footnote 17 Conversely, authoritarian regimes are viewed as closed ones if they allow a single party (or a single presidential candidate) to contest elections or do not have legislatures or elected executives at all.Footnote 18 By using both Svolik’s data and the DPI in identifying electoral authoritarian regimes, we cover the period of 1946–2012.Footnote 19 If legislative (executive) elections satisfy the conditions for multi-party competition in Parliament-based (presidential) systems, a country is seen as an electoral authoritarian regime.Footnote 20

Independent Variable

Our independent variable of interest concerns executive selection systems, that is, the difference between Parliament-based systems and Presidential systems. Following Roberts, our analysis is based on the SYSTEM variable in the World Bank’s DPI (Roberts 2015; Beck et al. 2001). This variable distinguishes three types of executive–legislative relations: (1) parliamentary, (2) assembly-elected president, and (3) Presidential systems. A country is coded parliamentary when the legislature elects the chief executive. A system with assembly-elected presidents refers to a government in which the assembly-elected chief executive (president) cannot be “easily recalled.”Footnote 21 A country is presidential when chief executives are elected directly by voters or an electoral college.Footnote 22 In the DPI, a country that has both a directly elected president and an assembly-elected prime minister is classified as presidential if the president can veto legislation and the legislature needs a supermajority to override the veto or if the president can appoint and dismiss the prime minister and dissolve parliament. In light of the discussion above, the DPIs’ SYSTEM variables (1) and (2) are parliamentary, while (3) incorporates both presidential and semi-presidential systems. We thereby create a dummy variable called “Parliament-based system,” coded 1 if the chief executive is elected by parliament (i.e., countries adopting parliamentary systems or having assembly-elected presidents) on the DPI and coded 0 (“Presidential”), if a country adopts a presidential or a semi-presidential system in the conventional terminology of comparative presidentialism literature. This dichotomous measurement effectively captures our conceptual distinction between Presidential- and Parliament-based autocracies because the classification centers on whether the constitution stipulates that the chief executive be chosen directly through elections. Although the DPI covers from 1975, we expand our coverage back to 1946, using various data sources (see Appendix E1). We also double-check the accuracy of our original coding as well as the DPI’s by referring to the V-Dem dataset, paying careful attention to differences in coding schemes between those two major datasets (Teorell and Lindberg 2019).Footnote 23

With the core variable, we include the following controls predicting democratic transitions based on the literature (all variables had a 1-year lag): logged GDP per capita (Boix and Stokes 2003), proportions of neighbors’ democracy and electoral autocracy (Brinks and Coppedge 2006), trade openness (Li and Reuveny 2003), natural resource wealth (Ross 2012), and the leader’s length of tenure.Footnote 24

Modeling Strategy

To test our hypothesis, we first estimate a probit model. The coefficient of the executive selection system variable is expected to be negative and statistically significant. Considering the likelihood that errors are correlated within each unit (country), we report country-clustered robust standard errors. As the dependent variable is binary, employing country-fixed effect models excludes countries that had never experienced regime change during the period, which might bias results. Therefore, instead of adopting country-level fixed-effects models, we introduce five regional dummies to consider regional-specific heterogeneous effects and include half-decade dummies to control for temporal-specific confounding factors.Footnote 25 We also include a time lapse regarding the last democratic transitions and three cubic splines to address the time-dependent nature of the dependent variable (Beck et al. 1998).

However, the most likely concern about employing the probit model to test the hypothesis is the possibility of endogeneity. First and foremost, as Pepinsky (2014) forcefully argued, the dictator’s choice of political institutions is influenced by unobservable heterogeneity that may affect both authoritarian institutions and regime change. For instance, it may be that cultures and history of cooperation within the ruling elites may influence both executive selection systems and the likelihood of democratic transitions. Durable electoral authoritarian regimes in Southeast Asia, for example, had a history of high cohesion among ruling elites due to their need to deal with insurgency in the independence era (Slater 2010). Such cooperative relationships within ruling elites, in turn, may affect the choice of executive selection systems on the one hand, and make democratic transition less likely on the other. Another possible confounding is to what extent autocrats have informal, personalized resources outside of formal institutions. If autocrats hold such personalized power bases, these informal resources may help autocrats stay in power on the one hand and encourage autocrats to select Presidential systems for the purpose of concentrating power in their hands. These historical, cultural, and informal developments are notoriously difficult to measure in the context of cross-national quantitative research, and adding observable variables as controls does not solve the problem.

Second, there might be reverse causality: dictators’ choice of executive selection systems could be influenced by the likelihood of regime transitions. There are two likely scenarios. On the one hand, dictators may decide to shift from a Presidential to a Parliament-based system with the expectation that doing so can extend their reign in the midst of increasing regime instability. This was indeed the case for President Park Chung Hee of South Korea, who moved to a Parliament-based system in the 1972 constitutional reform. Facing presidential term limits under the 1962 Constitution, Park created the National Conference for Unification, an electoral college that was charged with electing the president with no limits on reelection (Im 2011). On the other hand, authoritarian leaders may opt to shift from a Parliament-based to a Presidential system in order to expand their power in the middle of increasing regime stability. This was the case for many authoritarian leaders in Sub-Saharan Africa shortly after decolonization (Robinson and Torvik 2016). For example, Kwame Nkrumah, the first prime minister of Ghana after independence in 1957, changed the constitution and adopted a Presidential system in 1960. Later, in the 1964 constitutional amendment, he made himself “president for life” (Hadjor 1988).

To mitigate these endogeneity concerns, we employ IV estimators. The IV strategy is employed to demonstrate that an endogenous variable X (here, executive election systems) causes the dependent variable Y (here, democratic transitions). The IV strategy uses instruments Z that need to satisfy the following two conditions in order to identify the causal effect of X on Y. First, instrumental variables must be highly correlated with X and thus hold strong predictive power with respect to variations in the instrumented variable X. Second, instrumental variables must satisfy the exclusion restriction where Z’s influences on the dependent variable only go through the instrumented variable X. If Z violates this, then the instruments cannot be used to estimate the causal effects of X on Y.

This paper uses neighboring countries’ executive selection systems and their colonial legacies as instruments to estimate the causal effect of the country’s executive selection systems on democratic transitions. These two variables have strong theoretical and empirical reasons satisfying the two conditions discussed above.Footnote 26 For the first condition, the country’s choice of executive selection systems is strongly correlated with that of its neighbors. This comes mostly from the fact that political leaders often learn from institutional reforms in their neighborhoods and emulate neighbors’ institutions. For instance, after the collapse of the Soviet Union, post–Soviet electoral autocracies tended to mimic Russia’s constitutional arrangements and electoral laws. Such authoritarian diffusions of institutions were frequently observed in the region (Bader 2014).

The second good candidate of an instrumental variable is to what extent the country is surrounded by former British colonies. Countries dominated by the British Empire are very likely to have inherited parliamentary systems from the colonial master (Cheibub 2007: 150–151; Wiseman 1990: 21). As these studies suggest, other colonial origins, such as French and Spanish colonies, are not expected to work as strong predictors of the executive selection systems; thus, we focus on former British colonies. Importantly, since former British colonies tend to be geographically clustered, the country’s choice of an executive selection system tends to coincide with that of its neighbors. Moreover, if many neighbors have experienced British colonization, they are likely to adopt Parliament-based systems, which in turn influence the choice of the country’s executive selection system through learning and emulation.

As we show below, diagnostic tests strongly suggest that these variables are highly predictive of a country’s executive selection system. Instrumental variables’ coefficients in IV estimations’ first-stage models are all positive (as we expected), and they also well exceed standard benchmarks for strong instruments (Appendix F, Table F-1), strongly suggesting that consistent with the theoretical reasoning and anecdotes shown above, neighbors’ colonial origins and executive selection systems influence the country’s adoption of Parliament-based systems.

Regarding the second condition, the exclusion restriction, we think that it is improbable that neighbors’ colonial origins and their executive selection systems directly influence a given country’s political regime change. In fact, recent cross-national studies of democratization and autocratic politics have taken advantage of relationships with neighbors as instruments to mitigate possible endogeneity problems between political regimes and authoritarian institutions and/or development (Acemoglu et al. 2014; Miller 2015b; Knutsen et al. 2017). One possible concern about this approach is that if neighbors’ executive selection systems and colonial origins affect their political regimes, and also their regime changes diffuse to the country of interest, then the exclusion restriction is violated. Therefore, following recent studies addressing a similar issue, we include neighboring countries’ proportions of democracies and electoral autocracies as controls to block diffusion effects of democratization and other potential confounders (Miller 2015b). As a robustness check, we also control for foreign aid and the country’s colonial legacy to guard against possibilities that neighbors sharing colonial origins provide foreign aid to the target countryFootnote 27 and/or influence regime change via the target state’s colonial heritage. The results remain robust, and the diagnostics tests indicate that the instruments are valid even after including those additional variables as controls (for details, see Appendix F, Table F-2).

Our IV models first predict the likelihood of adopting Parliament-based systems in a country by including neighbors’ proportions of Parliament-based systems and former British colonies. Whether a country was a British colony was measured by the Correlates of War Colonial History Dataset (Hensel 2014), whereas countries are regarded as neighbors if their two closest points of their outer boundaries are within 950 km, following Gleditsch and Ward (2006). Then, in the second model, we perform a regression on democratic transitions regarding the instrumented executive selection systems dummy.

For IV estimation, we employ two methods. First, we use the IV probit model because the dependent variable is binary. As the endogenous regressor (i.e., executive selection systems) is also binary, we use Roodman’s conditional mixed process (CMP) regression with country-clustered robust standard errors, wherein we can flexibly construct simultaneous equation models, including variables with continuous, ordinal, and binary scales on both sides of each equation (Roodman 2011). Second, the CMP regression does not report diagnostic tests necessary to judge the instruments’ validity, such as an F test of excluded instruments and Hansen’s J statistic for overidentifying restrictions. Therefore, we also estimate an IV-linear probability model (via generalized method of moments, GMM) with the Driscoll-Kraay standard errors (DKSE). The DKSE allows us to consider serial correlations simultaneously within panel units (country), spatial correlation across units, and heteroscedasticity (Driscoll and Kraay 1998).Footnote 28

Statistical Results

Table 1 shows our main results. Using the non-IV probit estimation, models 1 and 2 examine the executive selection system’s impact on the democratic transition of electoral authoritarian regimes. The coefficients are negative but very small and become statistically insignificant after controlling for the basic confounding variables. As we discussed, however, the non-IV probit model does not consider the likely possibilities of omitted variables bias and reverse causality. Given that there are possible unobservable confounding factors and possibilities that electoral autocrats tend to change executive selection systems strategically when facing pressures for democratization, the negative impact of Parliament-based systems on regime change may be unreliable without addressing endogeneity.

Table 1 Parliament-based systems prevent democratic transitions from electoral authoritarianism
Full size table

To deal with endogeneity problems, models 3 and 4 estimate IV regressions. The models include all the controls, and they all indicate that electoral autocracies with Parliament-based systems are less likely to democratize. The results imply that the relationship between executive selection systems and democratization suffers from endogeneity problems and after considering both observable heterogeneity and reverse causality by the IV estimators, Parliament-based electoral autocracies become negatively associated with democratic transitions. The IV-GMM estimator in model 5 reports several diagnostic tests for the IV estimation’s validity.Footnote 29 In the first stage, instrumental variables have the expected associations. The weak identification test investigates whether the instruments explain a sufficient amount of variance in the endogenous variable. Here, we show the Kleibergen-Paap F statistic, which, as a rule of thumb, should be more than 10 for the instruments to be considered valid. The F statistic is 42.18, suggesting that the instruments effectively predict the instrumented variable (executive selection systems). Regarding the exclusion restriction, the overidentifying test (Hansen’s J statistic) does not reject the null hypothesis, suggesting that the instruments are valid in that they are not correlated with the error term.Footnote 30

Figure 1 shows the difference between the predicted probabilities of democratic transitions in Presidential- and Parliament-based systems (based on model 4). The probability of democratic transition the following year in an electoral authoritarian country with a Presidential system is approximately 17.1%. Conversely, the probability of an electoral authoritarian country with a Parliament-based system transitioning to democracy in the following year is approximately 0.84%, meaning that electoral authoritarian regimes with Parliament-based systems are 16.3% less likely to democratize than those with Presidential systems. Examples of stable, Parliament-based, authoritarian regimes in our dataset (1946–2012) include Botswana (1966–2012), Egypt (1980–2005), Malaysia (1974–2012), and Indonesia (1971–1998).

Fig. 1
figure 1

Predicted probabilities of democratic transitions from electoral authoritarianism (based on model 3)

Full size image

To ensure that these statistical results are robust, we investigate additional issues (Appendix B): (1) the analysis’s extension to the relationship between executive selection systems and autocratic breakdown (Table B-1), (2) potential heteroscedasticity within regions (Table B-2), (3) an alternative method to deal with time dependence (Table B-3), (4) inclusion of additional control variables (Table B-4), (5) exclusion of the leaders’ tenure variable (Table B-5), (6) alternative measures of multi-party competition in electoral autocracies (Table B-6), and (7) recoding the executive selection system variable in light of power relationships between the chief executive and legislature (Table B-7).Footnote 31 In all these robustness checks, the main results remain unchanged.

Testing Additional Empirical Implications

We have pointed to two mechanisms resulting in Parliament-based electoral autocracies being less likely to democratize. First, Parliament-based systems are more likely to incentivize ruling elites to engage in power-sharing through institutionalized ruling parties. To statistically test this empirical implication, we estimate the impact of executive selection systems on the emergence of an institutionalized ruling party. Here, we use Geddes and her colleagues’ party personalism index, which measures to what extent power is concentrated on the dictator vis-à-vis his support party (Geddes et al. 2018). Specifically, the variable measures whether the leader (1) makes access to office dependent on personal loyalty, (2) creates a new support party after seizing power, (3) controls appointments to the party executive committee, and (4) makes the party executive committee serve as a rubber stamp for his decisions. Aggregating these subcomponents by using an Item Response Theory technique, Geddes and her colleagues create an index of party personalism. As the party personalism variable is continuous, we run Prais-Winsten regressions with panel-corrected standard errors and country-fixed effects (Appendix Table C-1). Figure 2a demonstrates that in Presidential systems, ruling parties work less as a power-sharing institution than those in Parliament-based systems in a statistically significant way, which is consistent with our theoretical expectation.

Fig. 2
figure 2

Executive selection systems, party personalism, and electoral fraud

Full size image

The other mechanism occurs because, in parliamentary elections, autocrats can engage in gerrymandering and enjoy larger seat biases through malapportionment, as those elections are often divided into various constituencies. These indirect electoral manipulation techniques enable them to win elections with fewer votes and without extensive blatant electoral fraud. On the contrary, ceteris paribus, autocrats in Presidential systems are more likely to blatantly rig presidential elections because most have only one nationwide electoral district. One testable statement of this mechanism is that presidential elections in Presidential systems should be more likely to experience blatant electoral fraud. To test this, we use Hyde and Marinov’s National Elections in Democracy and Autocracy (NELDA) to operationalize blatant electoral fraud (Hyde and Marinov 2012). They provide a dichotomous assessment of whether there were significant concerns that elections would not be free and fair (NELDA 11), which is used to measure concerns of blatant fraud. We then use three dummy variables to test our theoretical expectations: parliamentary elections in Parliament-based systems, presidential elections in Presidential systems, and the other election types (i.e., presidential [legislative] elections in Parliament-based [presidential] systems). We set presidential elections in Presidential systems as the reference category to compare with parliamentary elections in Parliament-based systems. Using country-election year data, we adopt probit regressions with a series of relevant control variables explaining electoral fraud (Table C-2).Footnote 32 Figure 2b shows that in legislative elections under Parliament-based systems, only 35% of parliamentary elections are exposed to serious concerns regarding electoral malpractice.Footnote 33 Conversely, in Presidential systems, 62% of presidential elections are concerned with extensive electoral fraud. These results support our theoretical expectations.

Conclusion

Although many scholars have labeled the parliamentary system “virtuous” (Linz 1990a, 1990b; Lijphart 1991; Horowitz 1990), our analyses suggest that the parliamentary system may be a “peril” to a country’s democratic prospects (Linz 1990a), in that one of its defining features (its system of selecting the chief executive) helps prevent democratic transition from electoral authoritarianism. This occurs because the following are likely to result in a Parliament-based system: the ruling party becomes more institutionalized and can function as a power-sharing device among authoritarian elites, charismatic opposition leaders are less likely to emerge, and there are a wider range of options for electoral manipulation tactics, such as malapportionment and gerrymandering. We have examined whether electoral authoritarian regimes with Parliament-based systems tend to prevent democratization effectively as compared to those in Presidential systems. This analysis has been completed through several cross-national statistical analyses, which provide supporting evidence for our theoretical argument.

Additionally, the present study has some implications on the unresolved issue of determining the actual level of influence that political institutions have on democratic transitions. Conflicting observations and arguments have been made regarding this issue. For example, in some cases, autocrats have used the legislature to divide and conquer opposition parties (Lust-Okar 2004), but in other cases, they have failed to do so (Gandhi and Reuter 2013). Furthermore, scholars seem to agree that an institutionalized ruling party lengthens the survival of an authoritarian regime. Nonetheless, it is puzzling why some dictators have succeeded in institutionalizing the ruling party (e.g., Zimbabwe’s ZANU-PF and United Russia under Putin), whereas others have failed (e.g., Kyrgyzstan and Marcos’s Philippines). Similarly, some elections have triggered the ousting of autocrats (e.g., Georgia in 2003, Ukraine in 2004, and the Philippines in 1986), whereas others have contributed to prolonging authoritarian rule (Malaysia, Egypt, Kazakhstan, Russia, and Singapore). One way to reconcile these conflicting observations is to incorporate the dimension of the executive selection system into institutional analyses. In a Parliament-based system, legislatures tend to become a tool for developing an institutionalized ruling party and deftly manipulating election results (these abilities may be absent in Presidential systems). Our analyses suggest that the functions of these seemingly democratic institutions in autocracies may differ depending on the available executive selection methods.